Causality in Time Series - ClopiNet

More documents

Recommendations

Info

Guyon Statnikov Aliferismethods on some particular difficulties, and fostering the development of new algorithms.The results indicated that causal discovery from observational data is not animpossible task, but a very hard one and pointed to the need for further research andbenchmarks (Guyon et al., 2008). The Causal Explorer package (Aliferis et al., 2003),which we had made available to the participants and is downloadable as shareware,proved to be competitive and is a good starting point for researchers new to the field. Itis a Matlab toolkit supporting “local” causal discovery algorithms, efficient to discoverthe causal structure around a target variable, even for a large number of variables. Thealgorithms are based on structure learning from tests of conditional independence, asall the top ranking methods in this first challenge.The first challenge (Guyon et al., 2008) explored an important problem in causalmodeling, but is only one of many possible problem statements. The second challenge(Guyon et al., 2010) called “competition pot-luck” aimed at enlarging the scope ofcausal discovery algorithm evaluation by inviting members of the community to submittheir own problems and/or solve problems proposed by others. The challenge startedSeptember 15, 2008 and ended November 20, 2008, see http://www.causality.inf.ethz.ch/pot-luck.php. One task proposed by a participant drew a lot ofattention: the cause-effect pair task. The problem was to try to determine in pairs ofvariables (of known causal relationships), which one was the cause of the other. Thisproblem is hard for a lot of algorithms, which rely on the result of conditional independencetests of three or more variables. Yet the winners of the challenge succeeded inunraveling 8/8 correct causal directions (Zhang and Hyvärinen, 2009).Our planned challenge ExpDeCo (Experimental Design in Causal Discovery) willbenchmark methods of experimental design in application to causal modeling. The goalwill be to identify effective methods to unravel causal models, requiring a minimum ofexperimentation, using the Virtual Lab. A budget of virtual cash will be allocated toparticipants to “buy” the right to observe or manipulate certain variables, manipulationsbeing more expensive that observations. The participants will have to spend theirbudget optimally to make the best possible predictions on test data. This setup lendsitself to incorporating problems of relevance to development projects, in particular inmedicine and epidemiology where experimentation is difficult while developing newmethodology.We are planning another challenge called CoMSICo for “Causal Models for SystemIdentification and Control”, which is more ambitious in nature because it will perform acontinuous evaluation of causal models rather than separating training and test phase. Incontrast with ExpDeCo in which the organizers will provide test data with prescribedmanipulations to test the ability of the participants to make predictions of the consequencesof actions, in CoMSICo, the participants will be in charge of making theirown plan of action (policy) to optimize an overall objective (e.g., improve the life expectancyof a population, improve the GNP, etc.) and they will be judged directly withthis objective, on an on-going basis, with no distinction between “training” and “test”data. This challenge will also be via the Virtual Lab. The participants will be givenan initial amount of virtual cash, and, as previously, both actions and observations will136
Causality Workbenchhave a price. New in CoMSICo, virtual cash rewards will be given for achieving goodintermediate performance, which the participants will be allowed to re-invest to conductadditional experiments and improve their plan of action (policy). The winner willbe the participant ending up with the largest amount of virtual cash.6. ConclusionOur program of data exchange and benchmark proposes to challenge the research communitywith a wide variety of problems from many domains and focuses on realisticsettings. Causal discovery is a problem of fundamental and practical interest in manyareas of science and technology and there is a need for assisting policy making in allthese areas while reducing the costs of data collection and experimentation. Hence, theidentification of efficient techniques to solve causal problems will have a widespreadimpact. By choosing applications from a variety of domains and making connectionsbetween disciplines as varied as machine learning, causal discovery, experimental design,decision making, optimization, system identification, and control, we anticipatethat there will be a lot of cross-fertilization between different domains.AcknowledgmentsThis project is an activity of the Causality Workbench supported by the Pascal networkof excellence funded by the European Commission and by the U.S. National ScienceFoundation under Grant N0. ECCS-0725746. Any opinions, findings, and conclusionsor recommendations expressed in this material are those of the authors and do not necessarilyreflect the views of the National Science Foundation. We are very grateful toall the members of the causality workbench team for their contribution and in particularto our co-founders Constantin Aliferis, Greg Cooper, André Elisseeff, Jean-PhilippePellet, Peter Spirtes, and Alexander Statnikov.ReferencesC. F. Aliferis, I. Tsamardinos, A. Statnikov, and L.E. Brown. Causal explorer: A probabilisticnetwork learning toolkit for biomedical discovery. In 2003 InternationalConference on Mathematics and Engineering Techniques in Medicine and BiologicalSciences (METMBS), Las Vegas, Nevada, USA, June 23-26 2003. CSREA Press.Constantin Aliferis. A Temporal Representation and Reasoning Model for MedicalDecision-Support Systems. PhD thesis, University of Pittsburgh, 1998.C. Glymour and G.F. Cooper, editors. Computation, Causation, and Discovery. AAAIPress/The MIT Press, Menlo Park, California, Cambridge, Massachusetts, London,England, 1999.I. Guyon, C. Aliferis, G. Cooper, A. Elisseeff, J.-P. Pellet, P. Spirtes, and A. Statnikov.Design and analysis of the causation and prediction challenge. In JMLR W&CP,137
Page 1:
Causality in Time SeriesVolume 5:Ca
Page 5:
PrefaceThe following is a print ver
Page 9 and 10:
JMLR: Workshop and Conference Proce
Page 11 and 12:
Linking Granger Causality and the P
Page 13 and 14:
Page 15 and 16:
Page 17 and 18:
Page 19 and 20:
Page 21 and 22:
Page 23 and 24:
Page 25 and 26:
Page 27 and 28:
Page 29 and 30:
Page 31 and 32:
Page 33 and 34:
Page 35 and 36:
Page 37 and 38:
Page 39:
Page 42 and 43:
Popescu1. IntroductionCausality is
Page 44 and 45:
Popescugeneral terms, if we are pre
Page 46 and 47:
PopescuType III error prob. γ = P
Page 48 and 49:
Popescuwhich would correspond to a
Page 50 and 51:
Popescuvalued vector. A Data Genera
Page 52 and 53:
Popescuy i =K∑︁A k y i−k + Bu
Page 54 and 55:
Popescuy i =K∑︁A k y i−k + Bu
Page 56 and 57:
PopescuFigure 1: SVAR causality and
Page 58 and 59:
Popescu6. Spectral methods and phas
Page 60 and 61:
PopescuH GCj→i|u = logD i − log
Page 62 and 63:
Popescu8.1. The cardinal transform
Page 64 and 65:
Popescuy N,i = Bx N,i (37)y = (1
Page 66 and 67:
Popescu(a) Unmixed colored noise(b)
Page 68 and 69:
PopescuAs we can see in both Figure
Page 70 and 71:
Popescuβ > 0 y C,i =x N,i =⎡K∑
Page 72 and 73:
Popescupretation of information flo
Page 74 and 75:
Popescuit is suggested that a princ
Page 76 and 77:
PopescuA. N. Kolmogorov and A. N. S
Page 78 and 79:
PopescuH. White and X. Lu. Granger
Page 80 and 81:
PopescuTable 6: α vs. ΨN * → 50
Page 82 and 83:
Roebroeck Seth Valdes-Sosaincreasin
Page 84 and 85:
Roebroeck Seth Valdes-SosaFigure 1:
Page 86 and 87:
Roebroeck Seth Valdes-Sosasignal fr
Page 88 and 89:
Roebroeck Seth Valdes-Sosablood flo
Page 90 and 91:
Roebroeck Seth Valdes-Sosaat lower
Page 92 and 93:
Roebroeck Seth Valdes-SosaTable 1:
Page 94 and 95: Roebroeck Seth Valdes-Sosa1984) and
Page 96 and 97: Roebroeck Seth Valdes-Sosaanalysis
Page 98 and 99: Roebroeck Seth Valdes-Sosamarginal
Page 100 and 101: Roebroeck Seth Valdes-SosaThe initi
Page 102 and 103: Roebroeck Seth Valdes-Sosaobserved
Page 104 and 105: Roebroeck Seth Valdes-Sosaand corti
Page 106 and 107: Roebroeck Seth Valdes-SosaDaniel Co
Page 108 and 109: Roebroeck Seth Valdes-SosaM. Havlic
Page 110 and 111: Roebroeck Seth Valdes-SosaA. Roebro
Page 112 and 113: Roebroeck Seth Valdes-SosaV. Walsh
Page 114 and 115: Moneta Chlass Entner HoyerA traditi
Page 116 and 117: Moneta Chlass Entner Hoyerto the
Page 118 and 119: Moneta Chlass Entner Hoyersecond st
Page 120 and 121: Moneta Chlass Entner HoyerThe Wald
Page 122 and 123: Moneta Chlass Entner HoyerStep 3 Ap
Page 124 and 125: Moneta Chlass Entner Hoyerthe struc
Page 126 and 127: Moneta Chlass Entner Hoyerstricted
Page 128 and 129: Moneta Chlass Entner Hoyerwhere h 1
Page 130 and 131: Moneta Chlass Entner Hoyerlatter re
Page 132 and 133: Moneta Chlass Entner Hoyercausal mo
Page 134 and 135: Moneta Chlass Entner HoyerL. Baring
Page 136 and 137: Moneta Chlass Entner HoyerS. Johans
Page 138 and 139: Moneta Chlass Entner HoyerA. Yatche
Page 140 and 141: Guyon Statnikov Aliferisis extremel
Page 142 and 143: Guyon Statnikov Aliferis4. A Data R
Page 146 and 147: Guyon Statnikov Aliferisvolume 3, p
show all

Causality in Time Series - ClopiNet

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?